2 and ? 4 , will be significant if the predicted future OCF differs for firms using classification choices allowable under IFRS but not under U.S. GAAP. On one hand, we would expect the FLEX interaction coefficients to be positive because OCF using IFRS classification choices is higher than OCF using U.S. GAAP classifications. On the other hand, unlike the sales model, the independent variables are past cash flows and past accruals. Because past cash flows and past accruals capture the firm's classification choices in the prediction of future cash flows (using those same classification choices), these variables serve as controls for the classification choice also. In this case, the FLEX interaction coefficients will not be significant.
Table 1 presents our initial sample selection procedures. We select our sample of firms based on data availability in 2008 and then extend the sample to 2012, for a total sample period of 2005 to 2012. Footnote 16 We identify all nonfinancial firms in Compustat Global with key data items for all fiscal years from 2005 to 2008, including total assets, OCF, and market values. We exclude financial firms because prior research shows that the informativeness of these firms' cash flow statements differs from those of nonfinancial firms (Beatty et al. 2016). This selection procedure yields 2815 available firms.
Because databases do not accurately report cash flow classification, we hand collect the detailed cash flow items from the financial statements. For those countries with 100 available firms or less, we select 100% of the firms. For those countries with over 100 available firms, we select the greater of 100 firms or 30% of the firms with available data. Because of the large number of firms in the United Kingdom, we selected 15% (or 146) of total potential firms to collect the cash flow data. When sampling from the available population of firms within a country, we use stratified sampling, first ranking within country by industry and size (total assets) and then selecting firms. This selection procedure results in a potential sample of 1204 firms. Our final sample is reduced to 798 firms primarily because of non-accessible financial statements. The financial statements are either missing in Mergent Online (94% of cases) or written in languages other than English, German, or Danish. For the 798 sample firms, we collect all available data for the period from 2005 to 2012.
Sample selection and classification choices
Table 2 presents a description of the size and profitability of the 798 firms in the final sample and a comparison with other firms in the country that were excluded because of inaccessible financial statements. As expected, the 798 firms in the final sample are generally larger (and, on average, more profitable) than the firms that were excluded.
In our data collection, we encountered an unexpected absence of interest paid, consistent with a possible noncompliance issue with regard to disclosure of interest paid. Footnote 17 For 1347 observations, we could not locate interest paid or where it was classified on the statement of cash flows after searching the statement of cash flows and the financial statement footnotes. Footnote 18 These firms may not have paid interest or interest paid may have been immaterial. However, we confirm that 1305 (1325) observations had interest expense (long-term debt) in Compustat Global and thus likely paid interest. Based on our review of disclosures by other firms, we determine that, if the interest paid had been in the investing or financing sections, it would likely have appeared as a separate line in the section in the statement of cash flows. Footnote 19 Therefore we categorize these observations as reporting interest paid in operating for our analyses. This classification tends to understate the difference between IFRS and U.S. GAAP.